A novel burning state recognition method for rotary kilns is proposed. The existing methods are incompetent to extract distinguished and robust local shape features. As the data in each region of interest have different statistical properties, the convolutional independent subspace analysis networks are applied to extract local shape features of increasing complexity layer by layer from all the regions of interest firstly. Then, the bag of words model is used to extract global shape features. Finally, an integrated classifier based on random vector functional-link networks and fuzzy integral is designed. Experimental results show that the proposed method can improve the recognition accuracy of the burning states.